AI Paper Atlas

Research map / Detection and segmentation / CNN features

Forensics Adapter: Unleashing CLIP for Generalizable Face Forgery Detection

Xinjie Cui, Yuezun Li, Delong Zhu, Jiaran Zhou, Junyu Dong, Siwei Lyu

TPAMI 2026 · Detection and segmentation › CNN features · approach: Forgery detection · 9 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add patch-wise contrastive loss L2 (ablation)ForAda · CDF-v2 (frame-level)AUC: 0.822 -> 0.900Improved
add masked blending boundary loss L1 (ablation)ForAda · CDF-v2 (frame-level)AUC: 0.860 -> 0.900Improved
add sample-wise contrastive loss L3 (ablation)ForAda · DFDCP (frame-level)AUC: 0.893 -> 0.890Mixed
apply contrastive learning on both adapter and CLIP (ablation)ForAda · CDF-v2 (frame-level)AUC: 0.789 -> 0.900Improved
apply contrastive learning on CLIP side only (ablation)ForAda · CDF-v2 (frame-level)AUC: 0.789 -> 0.875Improved
increase adapter depth from 8 to 12 layers (ablation)ForAda · CDF-v1 (frame-level)AUC: 0.914 -> 0.877Hurt
raise fake-patch area threshold from 10% to 100% (ablation)ForAda · DFDC (frame-level)AUC: 0.843 -> 0.782Hurt
add binary boundary mask B to the boundary MSE loss (ablation)ForAda · CDF-v2 (frame-level)AUC: 0.893 -> 0.900Improved
add forgery-aware region prompts (text modality) over no text (ablation)ForAda++ · DFDC (frame-level)AUC: 0.838 -> 0.863Improved
use shared real/fake prompts instead of no text (ablation)ForAda++ · DFDC (frame-level)AUC: 0.838 -> 0.837No clear effect
add binary mask B (forged region) to ForAda++ loss (ablation)ForAda++ · DFDC (frame-level)AUC: 0.863 -> 0.837Hurt
add learnable prompt suffix of 8 tokens (ablation)ForAda++ · CDF-v2 (frame-level)AUC: 0.879 -> 0.903Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy